Machine Learning-Based Early Detection of Diabetic Retinopathy: A Comparative Study Using BP-ANN and SVM

Preventing vision loss and providing appropriate medical intervention are both made possible by early detection of diabetic retinopathy (DR). This study's overarching goal is to improve upon previous efforts in early DR detection by creating and testing a model that combines prior information with an enhanced Backpropagation Artificial Neural Network (BP-ANN). The model will then be compared to both the classic BP-ANN and Support Vector Machine (SVM) approaches. Retinal fundus images were gathered into a dataset. Essential characteristics of the retina, including the width and tortuosity of blood vessels, were semi-automatically retrieved and utilized as inputs for a priori knowledge. In addition to conventional BP-ANN and SVM models, we trained an enhanced BP-ANN model that leveraged these attributes. A 10-trial, 5-fold cross-validation procedure was used to assess the generalizability and robustness of the model. In terms of training efficiency, convergence speed, and classification accuracy, the experimental findings showed that the enhanced BP-ANN model outperformed both the conventional BP-ANN and SVM models. The results show that the improved BP-ANN could be a valuable tool for ocular diagnostics and confirm that including domain-specific features into neural networks can improve early DR diagnosis.